Skip to content

Latest commit

 

History

History
349 lines (277 loc) · 14.5 KB

reports.md

File metadata and controls

349 lines (277 loc) · 14.5 KB
Raw

Reports

The Report custom resource is used to manage the execution and status of reports. Metering produces reports derived from usage data sources which can be used in further analysis and filtering.

Report object

A single Report resource represents a report which is updated with new information according to a schedule. Reports with a spec.schedule field set are always running, and will track what time periods it has collected data for, ensuring that if Metering is shutdown or unavailable for an extended period of time, it will backfill the data starting where it left off. If the schedule is unset, then the Report will run once for the time specified by the reportingStart and reportingEnd. By default, reports will wait for ReportDataSources to have progressed in their import process to cover the report period being processed. If the report has a schedule, it will wait until the period currently being processed has been covered by the import process.

Example Report with a Schedule

The following example Report will contain information on every Pod's CPU requests, and will run every hour, adding the last hours worth of data each time it runs.

apiVersion: metering.openshift.io/v1
kind: Report
metadata:
  name: pod-cpu-request-hourly
spec:
  query: "pod-cpu-request"
  reportingStart: "2019-07-01T00:00:00Z"
  schedule:
    period: "hourly"
    hourly:
      minute: 0
      second: 0

Example Report without a Schedule (Run-Once)

The following example report will contain information on every Pod's CPU requests for all of July. After completion it does not run again.

apiVersion: metering.openshift.io/v1
kind: Report
metadata:
  name: pod-cpu-request-hourly
spec:
  query: "pod-cpu-request"
  reportingStart: "2019-07-01T00:00:00Z"
  reportingEnd: "2019-07-31T00:00:00Z"

query

Names the ReportQuery used to generate the report. The report query controls the schema of the report as well how the results are processed.

query is a required field.

Use kubectl to obtain a list of available ReportQuery objects:

$ kubectl -n $METERING_NAMESPACE get reportqueries
NAME                                         AGE
cluster-cpu-capacity                         23m
cluster-cpu-capacity-raw                     23m
cluster-cpu-usage                            23m
cluster-cpu-usage-raw                        23m
cluster-cpu-utilization                      23m
cluster-memory-capacity                      23m
cluster-memory-capacity-raw                  23m
cluster-memory-usage                         23m
cluster-memory-usage-raw                     23m
cluster-memory-utilization                   23m
cluster-persistentvolumeclaim-request        23m
namespace-cpu-request                        23m
namespace-cpu-usage                          23m
namespace-cpu-utilization                    23m
namespace-memory-request                     23m
namespace-memory-usage                       23m
namespace-memory-utilization                 23m
namespace-persistentvolumeclaim-request      23m
namespace-persistentvolumeclaim-usage        23m
node-cpu-allocatable                         23m
node-cpu-allocatable-raw                     23m
node-cpu-capacity                            23m
node-cpu-capacity-raw                        23m
node-cpu-utilization                         23m
node-memory-allocatable                      23m
node-memory-allocatable-raw                  23m
node-memory-capacity                         23m
node-memory-capacity-raw                     23m
node-memory-utilization                      23m
persistentvolumeclaim-capacity               23m
persistentvolumeclaim-capacity-raw           23m
persistentvolumeclaim-phase-raw              23m
persistentvolumeclaim-request                23m
persistentvolumeclaim-request-raw            23m
persistentvolumeclaim-usage                  23m
persistentvolumeclaim-usage-raw              23m
persistentvolumeclaim-usage-with-phase-raw   23m
pod-cpu-request                              23m
pod-cpu-request-raw                          23m
pod-cpu-usage                                23m
pod-cpu-usage-raw                            23m
pod-memory-request                           23m
pod-memory-request-raw                       23m
pod-memory-usage                             23m
pod-memory-usage-raw                         23m

ReportQueries with the -raw suffix are used by other ReportQueries to build more complex queries, and should not be used directly for reports.

namespace- prefixed queries aggregate Pod CPU/memory requests by namespace, providing a list of namespaces and their overall usage based on resource requests.

pod- prefixed queries are similar to 'namespace-' prefixed, but aggregate information by Pod, rather than namespace. These queries include the Pod's namespace and node.

node- prefixed queries return information about each node's total available resources.

aws- prefixed queries are specific to AWS. Queries suffixed with -aws return the same data as queries of the same name without the suffix, and correlate usage with the EC2 billing data.

The aws-ec2-billing-data report is used by other queries, and should not be used as a standalone report. The aws-ec2-cluster-cost report provides a total cost based on the nodes included in the cluster, and the sum of their costs for the time period being reported on.

For a complete list of fields each report query produces, use kubectl to get the object as YAML, and check the columns field:

$ kubectl -n $METERING_NAMESPACE get reportqueries namespace-memory-request -o yaml

apiVersion: metering.openshift.io/v1
kind: ReportQuery
metadata:
  name: namespace-memory-request
  labels:
    operator-metering: "true"
spec:
  columns:
  - name: period_start
    type: timestamp
    unit: date
  - name: period_end
    type: timestamp
    unit: date
  - name: namespace
    type: varchar
    unit: kubernetes_namespace
  - name: pod_request_memory_byte_seconds
    type: double
    unit: byte_seconds

schedule

The schedule block defines when the report runs. The main fields in the schedule section are period, and then depending on the value of period, the fields hourly, daily, weekly and monthly allow you to fine-tune when the report runs.

For example, if period is set to weekly, you can add a weekly key to the schedule block. The following example will run once a week on Wednesday, at 1 PM.

...
  schedule:
    period: "weekly"
    weekly:
      dayOfWeek: "wednesday"
      hour: 13

period

Valid values of period are listed below, and the options available to set for a given period are also listed.

  • hourly
    • minute
    • second
  • daily
    • hour
    • minute
    • second
  • weekly
    • dayOfWeek
    • hour
    • minute
    • second
  • monthly
    • dayOfMonth
    • hour
    • minute
    • second
  • cron
    • expression

Generally, the hour, minute, second fields control when in the day the report runs, and dayOfWeek/dayOfMonth control what day of the week, or day of month the report runs on, if it's a weekly or monthly report period.

For each of these fields, there is a range of valid values:

  • hour is an integer value between 0-23.
  • minute is an integer value between 0-59.
  • second is an integer value between 0-59.
  • dayOfWeek is a string value that expects the day of the week (spelled out).
  • dayOfMonth is an integer value between 1-31.

For cron periods, normal cron expressions are valid:

  • expression: "*/5 * * * *"

reportingStart

To support running a Report against existing data, you can set the spec.reportingStart field to a RFC3339 timestamp to tell the Report to run according to its schedule starting from reportingStart rather than the current time. One important thing to understand is that this will result in the reporting-operator running many queries in succession for each interval in the schedule that's between the reportingStart time and the current time. This could be thousands of queries if the period is less than daily and the reportingStart is more than a few months back. If reportingStart is left unset, the Report will run at the next full reportingPeriod after the time the report is created.

As an example of how to use this field, if you had data already collected dating back to January 1st, 2019 which you wanted to be included in your Report, you could create a report with the following values:

apiVersion: metering.openshift.io/v1
kind: Report
metadata:
  name: pod-cpu-request-hourly
spec:
  query: "pod-cpu-request"
  schedule:
    period: "hourly"
  reportingStart: "2019-01-01T00:00:00Z"

reportingEnd

To configure a Report to only run until a specified time, you can set the spec.reportingEnd field to an RFC3339 timestamp. The value of this field will cause the Report to stop running on its schedule after it has finished generating reporting data for the period covered from its start time until reportingEnd. Because a schedule will most likely not align with reportingEnd, the last period in the schedule will be shortened to end at the specified reportingEnd time. If left unset, then the Report will run forever, or until a reportingEnd is set on the Report.

For example, if you wanted to create a report that runs once a week for the month of July:

apiVersion: metering.openshift.io/v1
kind: Report
metadata:
  name: pod-cpu-request-hourly
spec:
  query: "pod-cpu-request"
  schedule:
    period: "weekly"
  reportingStart: "2019-07-01T00:00:00Z"
  reportingEnd: "2019-07-31T00:00:00Z"

expiration

Add the expiration field to set a retention period on a scheduled metering Report. You can avoid manually removing the Report by setting the expiration duration value. The retention period is equal to the Report creationDate plus the expiration duration. The Report is removed from the cluster at the end of the retention period if no other Reports or ReportQueries depend on the expiring Report. Deleting the Report from the cluster can take several minutes.

Setting the expiration field is not recommended for roll-up or aggregated reports. If a Report is depended upon by other Reports or ReportQueries, then the Report is not removed at the end of the retention period. You can view the reporting-operator logs at debug level for the timing output around a Report retention decision.

For example, the following scheduled Report is deleted 30 minutes after the metadata.creationDate of the Report:

apiVersion: metering.openshift.io/v1
kind: Report
metadata:
  name: pod-cpu-request-hourly
spec:
  query: "pod-cpu-request"
  schedule:
    period: "weekly"
  reportingStart: "2020-09-01T00:00:00Z"
  expiration: "30m"

Valid time units for the expiration duration are ns, us (or µs), ms, s, m, and h.

The expiration retention period for a Report is not precise and works on the order of several minutes, not nanoseconds.

runImmediately

When runImmediately is set to true, the report will be run immediately. This behavior ensures that the report is immediately processed and queued without requiring additional scheduling parameters.

Note: When runImmediately is set to true you must set a reportingEnd and reportingStart value.

Inputs

The spec.inputs field of a Report can be used to override or set values defined in a ReportQuery's spec.inputs field.

It is a list of name-value pairs:

spec:
  inputs:
  - name: "NamespaceCPUUsageReportName"
    value: "namespace-cpu-usage-hourly"

The name of an input must exist in the ReportQuery's inputs list. The value of the input must be the correct type for the input's type.

For an example of how this can be used, see it in action in a roll-up report. For more details on how inputs can be specified read the Specifying Inputs section of the ReportQueries documentation.

Roll-up Reports

Report data is stored in the database much like metrics themselves, and can thus be used in aggregated or roll-up reports. A simple use case for a roll-up report is to spread the time required to produce a report over a longer period of time: instead of requiring a monthly report to query and add all data over an entire month, the task can be split into daily reports that each run over a thirtieth of the data.

A custom roll-up report requires a custom report query. The ReportQuery template processor provides a function: reportTableName that can get the necessary table name from a report name.

Below is a snippet taken from a built-in query:

# Taken from pod-cpu.yaml
spec:
...
  inputs:
  - name: ReportingStart
    type: time
  - name: ReportingEnd
    type: time
  - name: NamespaceCPUUsageReportName
    type: Report
  - name: PodCpuUsageRawDataSourceName
    type: ReportDataSource
    default: pod-cpu-usage-raw
...

  query: |
...
    {|- if .Report.Inputs.NamespaceCPUUsageReportName |}
      namespace,
      sum(pod_usage_cpu_core_seconds) as pod_usage_cpu_core_seconds
    FROM {| .Report.Inputs.NamespaceCPUUsageReportName | reportTableName |}
...
# aggregated-report.yaml
spec:
  query: "namespace-cpu-usage"
  inputs:
  - name: "NamespaceCPUUsageReportName"
    value: "namespace-cpu-usage-hourly"

For more information on setting up a roll-up report, see the roll-up report guide.

Report Status

The execution of a scheduled report can be tracked using its status field. Any errors occurring during the preparation of a report will be recorded here.

The status field of a Report currently has two fields:

  • conditions: Conditions is a list of conditions, each of which have a type, status, reason, and message field. Possible values of a condition's type field are Running and Failure, indicating the current state of the scheduled report. The reason indicates why its condition is in its current state with the status being either true, false or unknown. The message provides a human readable indicating why the condition is in the current state. For detailed information on the reason values see pkg/apis/metering/v1/util/report_util.go.
  • lastReportTime: Indicates the time Metering has collected data up to.